Move the world out of /zfs/ and seperate out module build tree
[zfs.git] / module / zfs / dsl_pool.c
diff --git a/module/zfs/dsl_pool.c b/module/zfs/dsl_pool.c
new file mode 100644 (file)
index 0000000..dacc57c
--- /dev/null
@@ -0,0 +1,613 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#include <sys/dsl_pool.h>
+#include <sys/dsl_dataset.h>
+#include <sys/dsl_dir.h>
+#include <sys/dsl_synctask.h>
+#include <sys/dmu_tx.h>
+#include <sys/dmu_objset.h>
+#include <sys/arc.h>
+#include <sys/zap.h>
+#include <sys/zio.h>
+#include <sys/zfs_context.h>
+#include <sys/fs/zfs.h>
+#include <sys/zfs_znode.h>
+#include <sys/spa_impl.h>
+
+int zfs_no_write_throttle = 0;
+int zfs_write_limit_shift = 3;                 /* 1/8th of physical memory */
+int zfs_txg_synctime = 5;                      /* target secs to sync a txg */
+
+uint64_t zfs_write_limit_min = 32 << 20;       /* min write limit is 32MB */
+uint64_t zfs_write_limit_max = 0;              /* max data payload per txg */
+uint64_t zfs_write_limit_inflated = 0;
+uint64_t zfs_write_limit_override = 0;
+
+kmutex_t zfs_write_limit_lock;
+
+static pgcnt_t old_physmem = 0;
+
+static int
+dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
+{
+       uint64_t obj;
+       int err;
+
+       err = zap_lookup(dp->dp_meta_objset,
+           dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
+           name, sizeof (obj), 1, &obj);
+       if (err)
+               return (err);
+
+       return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
+}
+
+static dsl_pool_t *
+dsl_pool_open_impl(spa_t *spa, uint64_t txg)
+{
+       dsl_pool_t *dp;
+       blkptr_t *bp = spa_get_rootblkptr(spa);
+
+       dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
+       dp->dp_spa = spa;
+       dp->dp_meta_rootbp = *bp;
+       rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
+       dp->dp_write_limit = zfs_write_limit_min;
+       txg_init(dp, txg);
+
+       txg_list_create(&dp->dp_dirty_datasets,
+           offsetof(dsl_dataset_t, ds_dirty_link));
+       txg_list_create(&dp->dp_dirty_dirs,
+           offsetof(dsl_dir_t, dd_dirty_link));
+       txg_list_create(&dp->dp_sync_tasks,
+           offsetof(dsl_sync_task_group_t, dstg_node));
+       list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
+           offsetof(dsl_dataset_t, ds_synced_link));
+
+       mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
+       mutex_init(&dp->dp_scrub_cancel_lock, NULL, MUTEX_DEFAULT, NULL);
+
+       return (dp);
+}
+
+int
+dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
+{
+       int err;
+       dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
+       dsl_dir_t *dd;
+       dsl_dataset_t *ds;
+       objset_impl_t *osi;
+
+       rw_enter(&dp->dp_config_rwlock, RW_WRITER);
+       err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
+       if (err)
+               goto out;
+       dp->dp_meta_objset = &osi->os;
+
+       err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+           DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
+           &dp->dp_root_dir_obj);
+       if (err)
+               goto out;
+
+       err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
+           NULL, dp, &dp->dp_root_dir);
+       if (err)
+               goto out;
+
+       err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
+       if (err)
+               goto out;
+
+       if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
+               err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
+               if (err)
+                       goto out;
+               err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
+                   FTAG, &ds);
+               if (err)
+                       goto out;
+               err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
+                   dp, &dp->dp_origin_snap);
+               if (err)
+                       goto out;
+               dsl_dataset_rele(ds, FTAG);
+               dsl_dir_close(dd, dp);
+       }
+
+       /* get scrub status */
+       err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+           DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1,
+           &dp->dp_scrub_func);
+       if (err == 0) {
+               err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+                   DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1,
+                   &dp->dp_scrub_queue_obj);
+               if (err)
+                       goto out;
+               err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+                   DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1,
+                   &dp->dp_scrub_min_txg);
+               if (err)
+                       goto out;
+               err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+                   DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1,
+                   &dp->dp_scrub_max_txg);
+               if (err)
+                       goto out;
+               err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+                   DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
+                   &dp->dp_scrub_bookmark);
+               if (err)
+                       goto out;
+               err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+                   DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
+                   &spa->spa_scrub_errors);
+               if (err)
+                       goto out;
+               if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) {
+                       /*
+                        * A new-type scrub was in progress on an old
+                        * pool.  Restart from the beginning, since the
+                        * old software may have changed the pool in the
+                        * meantime.
+                        */
+                       dsl_pool_scrub_restart(dp);
+               }
+       } else {
+               /*
+                * It's OK if there is no scrub in progress (and if
+                * there was an I/O error, ignore it).
+                */
+               err = 0;
+       }
+
+out:
+       rw_exit(&dp->dp_config_rwlock);
+       if (err)
+               dsl_pool_close(dp);
+       else
+               *dpp = dp;
+
+       return (err);
+}
+
+void
+dsl_pool_close(dsl_pool_t *dp)
+{
+       /* drop our references from dsl_pool_open() */
+
+       /*
+        * Since we held the origin_snap from "syncing" context (which
+        * includes pool-opening context), it actually only got a "ref"
+        * and not a hold, so just drop that here.
+        */
+       if (dp->dp_origin_snap)
+               dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
+       if (dp->dp_mos_dir)
+               dsl_dir_close(dp->dp_mos_dir, dp);
+       if (dp->dp_root_dir)
+               dsl_dir_close(dp->dp_root_dir, dp);
+
+       /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
+       if (dp->dp_meta_objset)
+               dmu_objset_evict(NULL, dp->dp_meta_objset->os);
+
+       txg_list_destroy(&dp->dp_dirty_datasets);
+       txg_list_destroy(&dp->dp_dirty_dirs);
+       list_destroy(&dp->dp_synced_datasets);
+
+       arc_flush(dp->dp_spa);
+       txg_fini(dp);
+       rw_destroy(&dp->dp_config_rwlock);
+       mutex_destroy(&dp->dp_lock);
+       mutex_destroy(&dp->dp_scrub_cancel_lock);
+       if (dp->dp_blkstats)
+               kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
+       kmem_free(dp, sizeof (dsl_pool_t));
+}
+
+dsl_pool_t *
+dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
+{
+       int err;
+       dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
+       dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
+       objset_impl_t *osip;
+       dsl_dataset_t *ds;
+       uint64_t dsobj;
+
+       /* create and open the MOS (meta-objset) */
+       dp->dp_meta_objset = &dmu_objset_create_impl(spa,
+           NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
+
+       /* create the pool directory */
+       err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+           DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
+       ASSERT3U(err, ==, 0);
+
+       /* create and open the root dir */
+       dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
+       VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
+           NULL, dp, &dp->dp_root_dir));
+
+       /* create and open the meta-objset dir */
+       (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
+       VERIFY(0 == dsl_pool_open_special_dir(dp,
+           MOS_DIR_NAME, &dp->dp_mos_dir));
+
+       if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
+               dsl_pool_create_origin(dp, tx);
+
+       /* create the root dataset */
+       dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
+
+       /* create the root objset */
+       VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
+       osip = dmu_objset_create_impl(dp->dp_spa, ds,
+           dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
+#ifdef _KERNEL
+       zfs_create_fs(&osip->os, kcred, zplprops, tx);
+#endif
+       dsl_dataset_rele(ds, FTAG);
+
+       dmu_tx_commit(tx);
+
+       return (dp);
+}
+
+void
+dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
+{
+       zio_t *zio;
+       dmu_tx_t *tx;
+       dsl_dir_t *dd;
+       dsl_dataset_t *ds;
+       dsl_sync_task_group_t *dstg;
+       objset_impl_t *mosi = dp->dp_meta_objset->os;
+       hrtime_t start, write_time;
+       uint64_t data_written;
+       int err;
+
+       tx = dmu_tx_create_assigned(dp, txg);
+
+       dp->dp_read_overhead = 0;
+       zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
+       while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
+               if (!list_link_active(&ds->ds_synced_link))
+                       list_insert_tail(&dp->dp_synced_datasets, ds);
+               else
+                       dmu_buf_rele(ds->ds_dbuf, ds);
+               dsl_dataset_sync(ds, zio, tx);
+       }
+       DTRACE_PROBE(pool_sync__1setup);
+
+       start = gethrtime();
+       err = zio_wait(zio);
+       write_time = gethrtime() - start;
+       ASSERT(err == 0);
+       DTRACE_PROBE(pool_sync__2rootzio);
+
+       while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
+               dsl_sync_task_group_sync(dstg, tx);
+       DTRACE_PROBE(pool_sync__3task);
+
+       start = gethrtime();
+       while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
+               dsl_dir_sync(dd, tx);
+       write_time += gethrtime() - start;
+
+       if (spa_sync_pass(dp->dp_spa) == 1)
+               dsl_pool_scrub_sync(dp, tx);
+
+       start = gethrtime();
+       if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
+           list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
+               zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
+               dmu_objset_sync(mosi, zio, tx);
+               err = zio_wait(zio);
+               ASSERT(err == 0);
+               dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
+               spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
+       }
+       write_time += gethrtime() - start;
+       DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
+           hrtime_t, dp->dp_read_overhead);
+       write_time -= dp->dp_read_overhead;
+
+       dmu_tx_commit(tx);
+
+       data_written = dp->dp_space_towrite[txg & TXG_MASK];
+       dp->dp_space_towrite[txg & TXG_MASK] = 0;
+       ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
+
+       /*
+        * If the write limit max has not been explicitly set, set it
+        * to a fraction of available physical memory (default 1/8th).
+        * Note that we must inflate the limit because the spa
+        * inflates write sizes to account for data replication.
+        * Check this each sync phase to catch changing memory size.
+        */
+       if (physmem != old_physmem && zfs_write_limit_shift) {
+               mutex_enter(&zfs_write_limit_lock);
+               old_physmem = physmem;
+               zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
+               zfs_write_limit_inflated = MAX(zfs_write_limit_min,
+                   spa_get_asize(dp->dp_spa, zfs_write_limit_max));
+               mutex_exit(&zfs_write_limit_lock);
+       }
+
+       /*
+        * Attempt to keep the sync time consistent by adjusting the
+        * amount of write traffic allowed into each transaction group.
+        * Weight the throughput calculation towards the current value:
+        *      thru = 3/4 old_thru + 1/4 new_thru
+        */
+       ASSERT(zfs_write_limit_min > 0);
+       if (data_written > zfs_write_limit_min / 8 && write_time > 0) {
+               uint64_t throughput = (data_written * NANOSEC) / write_time;
+               if (dp->dp_throughput)
+                       dp->dp_throughput = throughput / 4 +
+                           3 * dp->dp_throughput / 4;
+               else
+                       dp->dp_throughput = throughput;
+               dp->dp_write_limit = MIN(zfs_write_limit_inflated,
+                   MAX(zfs_write_limit_min,
+                   dp->dp_throughput * zfs_txg_synctime));
+       }
+}
+
+void
+dsl_pool_zil_clean(dsl_pool_t *dp)
+{
+       dsl_dataset_t *ds;
+
+       while (ds = list_head(&dp->dp_synced_datasets)) {
+               list_remove(&dp->dp_synced_datasets, ds);
+               ASSERT(ds->ds_user_ptr != NULL);
+               zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
+               dmu_buf_rele(ds->ds_dbuf, ds);
+       }
+}
+
+/*
+ * TRUE if the current thread is the tx_sync_thread or if we
+ * are being called from SPA context during pool initialization.
+ */
+int
+dsl_pool_sync_context(dsl_pool_t *dp)
+{
+       return (curthread == dp->dp_tx.tx_sync_thread ||
+           spa_get_dsl(dp->dp_spa) == NULL);
+}
+
+uint64_t
+dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
+{
+       uint64_t space, resv;
+
+       /*
+        * Reserve about 1.6% (1/64), or at least 32MB, for allocation
+        * efficiency.
+        * XXX The intent log is not accounted for, so it must fit
+        * within this slop.
+        *
+        * If we're trying to assess whether it's OK to do a free,
+        * cut the reservation in half to allow forward progress
+        * (e.g. make it possible to rm(1) files from a full pool).
+        */
+       space = spa_get_dspace(dp->dp_spa);
+       resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
+       if (netfree)
+               resv >>= 1;
+
+       return (space - resv);
+}
+
+int
+dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
+{
+       uint64_t reserved = 0;
+       uint64_t write_limit = (zfs_write_limit_override ?
+           zfs_write_limit_override : dp->dp_write_limit);
+
+       if (zfs_no_write_throttle) {
+               atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
+                   space);
+               return (0);
+       }
+
+       /*
+        * Check to see if we have exceeded the maximum allowed IO for
+        * this transaction group.  We can do this without locks since
+        * a little slop here is ok.  Note that we do the reserved check
+        * with only half the requested reserve: this is because the
+        * reserve requests are worst-case, and we really don't want to
+        * throttle based off of worst-case estimates.
+        */
+       if (write_limit > 0) {
+               reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
+                   + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
+
+               if (reserved && reserved > write_limit)
+                       return (ERESTART);
+       }
+
+       atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
+
+       /*
+        * If this transaction group is over 7/8ths capacity, delay
+        * the caller 1 clock tick.  This will slow down the "fill"
+        * rate until the sync process can catch up with us.
+        */
+       if (reserved && reserved > (write_limit - (write_limit >> 3)))
+               txg_delay(dp, tx->tx_txg, 1);
+
+       return (0);
+}
+
+void
+dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
+{
+       ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
+       atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
+}
+
+void
+dsl_pool_memory_pressure(dsl_pool_t *dp)
+{
+       uint64_t space_inuse = 0;
+       int i;
+
+       if (dp->dp_write_limit == zfs_write_limit_min)
+               return;
+
+       for (i = 0; i < TXG_SIZE; i++) {
+               space_inuse += dp->dp_space_towrite[i];
+               space_inuse += dp->dp_tempreserved[i];
+       }
+       dp->dp_write_limit = MAX(zfs_write_limit_min,
+           MIN(dp->dp_write_limit, space_inuse / 4));
+}
+
+void
+dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
+{
+       if (space > 0) {
+               mutex_enter(&dp->dp_lock);
+               dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
+               mutex_exit(&dp->dp_lock);
+       }
+}
+
+/* ARGSUSED */
+static int
+upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
+{
+       dmu_tx_t *tx = arg;
+       dsl_dataset_t *ds, *prev = NULL;
+       int err;
+       dsl_pool_t *dp = spa_get_dsl(spa);
+
+       err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
+       if (err)
+               return (err);
+
+       while (ds->ds_phys->ds_prev_snap_obj != 0) {
+               err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
+                   FTAG, &prev);
+               if (err) {
+                       dsl_dataset_rele(ds, FTAG);
+                       return (err);
+               }
+
+               if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
+                       break;
+               dsl_dataset_rele(ds, FTAG);
+               ds = prev;
+               prev = NULL;
+       }
+
+       if (prev == NULL) {
+               prev = dp->dp_origin_snap;
+
+               /*
+                * The $ORIGIN can't have any data, or the accounting
+                * will be wrong.
+                */
+               ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
+
+               /* The origin doesn't get attached to itself */
+               if (ds->ds_object == prev->ds_object) {
+                       dsl_dataset_rele(ds, FTAG);
+                       return (0);
+               }
+
+               dmu_buf_will_dirty(ds->ds_dbuf, tx);
+               ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
+               ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
+
+               dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
+               ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
+
+               dmu_buf_will_dirty(prev->ds_dbuf, tx);
+               prev->ds_phys->ds_num_children++;
+
+               if (ds->ds_phys->ds_next_snap_obj == 0) {
+                       ASSERT(ds->ds_prev == NULL);
+                       VERIFY(0 == dsl_dataset_hold_obj(dp,
+                           ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
+               }
+       }
+
+       ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
+       ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
+
+       if (prev->ds_phys->ds_next_clones_obj == 0) {
+               prev->ds_phys->ds_next_clones_obj =
+                   zap_create(dp->dp_meta_objset,
+                   DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
+       }
+       VERIFY(0 == zap_add_int(dp->dp_meta_objset,
+           prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
+
+       dsl_dataset_rele(ds, FTAG);
+       if (prev != dp->dp_origin_snap)
+               dsl_dataset_rele(prev, FTAG);
+       return (0);
+}
+
+void
+dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
+{
+       ASSERT(dmu_tx_is_syncing(tx));
+       ASSERT(dp->dp_origin_snap != NULL);
+
+       (void) dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
+           tx, DS_FIND_CHILDREN);
+}
+
+void
+dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
+{
+       uint64_t dsobj;
+       dsl_dataset_t *ds;
+
+       ASSERT(dmu_tx_is_syncing(tx));
+       ASSERT(dp->dp_origin_snap == NULL);
+
+       /* create the origin dir, ds, & snap-ds */
+       rw_enter(&dp->dp_config_rwlock, RW_WRITER);
+       dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
+           NULL, 0, kcred, tx);
+       VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
+       dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, kcred, tx);
+       VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
+           dp, &dp->dp_origin_snap));
+       dsl_dataset_rele(ds, FTAG);
+       rw_exit(&dp->dp_config_rwlock);
+}